Decentralized control for multi-robot systems with event-triggered communications

Multi-robot systems (MRSs) have been widely studied due to their features as flexibility and intrinsic resilience. Classical approaches to multi-robot control are based on centralized architectures, which are very effective in terms of performance but require a central unit and/or complete communication among all robots. This feature might be difficult to meet in practical scenarios or might be undesirable when mobile robots are involved. Therefore, recent studies have been mainly focused on decentralized or distributed systems, as they are better scalable and safer than centralized ones. Decentralized algorithms applied to networked robots should find out control and communication problems such as controlling the MRS centroid and the formation. This thesis deals with the simulation of a decentralized centroid and formation control, based on a distributed controller-observer scheme for a team of mobile agents (robots). In this context, each robot is characterized by a single-integrator dynamics x ̇=u, where x is the state and u is the input vector. A local observer is used by each agent to estimate the collective state of the system, while the distributed controller is in charge of desired centroid and formation tracking. This design leads to relevant advantages with respect to other layouts: - each agent evaluates the state of the whole system, which can be exploited to achieve other goals (e.g., fault detection); - estimation of the whole state results in a strong (i.e., exponential) convergence of the estimation and tracking errors. To reduce the exchange of information, each robot can communicate only with its neighbours. In the thesis, a multi-robot system characterized by a fixed and strongly connected directed communication graph is considered. Furthermore, in order to decrease the exchange of information between the robotic agents, an event-triggered control strategy is devised, where each agent communicates with its neighbours only when a triggering condition on the estimation error is verified. Two different triggering conditions are implemented: one based on a Lyapunov’s function for the estimation error and the other based on the dynamics of the estimation error. The triggering conditions are verified in simulation and their performance are compared. Matlab and Simulink have been used to perform the simulations.